【作者】 许兆新；

【导师】 郝燕玲；

【作者基本信息】 哈尔滨工程大学 ， 导航、制导与控制， 2003， 博士

【摘要】 从大量数据中抽取出有价值的信息和运行规律的数据挖掘技术在数据库技术发展了30余年的今天，无论是对商业经营还是工企决策，都有极其重要的经济价值和理论意义，但此新兴领域还有许多尚待解决的问题，还处于创新者提出概念研究者去解决问题的阶段。目前市场上已有多种适用商业模式的通用数据挖掘系统，但实际上这些系统极其不好用，只有那些非常熟悉数据挖掘技术的人才能够理解和使用它们，而普通用户是很难应用这些技术来解决自己的问题。 原因在于传统的数据挖掘模型缺乏充分表达挖掘目标的手段和必要的交互性且其评价标准单一，不能满足当前实际应用的需要。尽管通过设计高效算法及优化策略，会在某种程度上缓解这一矛盾，但并不能真正解决这一问题。本文以面向用户为原则，试图从数据挖掘系统的整体结构的高度来探讨解决这一问题的方法，其研究重点是提出基于元知识的数据挖掘系统模式并设计灵活、实用的算法。 本文的目的是赋予数据挖掘系统三个重要属性：交互性，可控性，易用性。由于用户不了解数据挖掘领域，他不知道要挖掘的是哪种类型的信息及如何正确表达系统需要他表达的思想，甚至不可能了解面对的数据库，元知识描述领域知识使得在这些方面能给与用户最大的帮助，本文给出了符合元数据交换标准的数据挖掘用元知识的内容、组织方式和应用模型，提出了基于元知识的数据挖掘系统体系结构，使数据库与数据挖掘系统真正融合，使领域知识在组织、表达、使用上有依据，为指导挖掘过程提供了手段，使数据量规模化有依据，数据源多样化有规范、可控制。 关联规则是数据挖掘研究的重要内容，高效的算法是数据挖掘的重要组成部分，而且面对复杂的多方面的需求，应提供针对各种因素的各具特色的互补的多种算法，本文详尽地研究了基于约束的关联规则挖掘算法的思想，深入地研究了约束的分类、确定其可利用的性质，在此基础上提出了针对联合约束中前驱和后继重叠时的改进策略和复杂聚集约束的转换方法。本文还给出了支持度匹配模式、0支持度约束等概念及模式匹配原则，引进枚举树确定模式最小支持度，设计了支持可变支持度约束的算法。哈尔滨工程大学博士学位论文元知识为灵活实现上述算法提供了支持手段。 最后，提出了利用信息增益选择约束的方法，给出了基于元知识的关联规则挖掘系统的操作流程，探讨了可视化的功能实现问题，提出了在结果评价阶段将评价规则作为元知识统一到挖掘过程中的思想，以利于提高全系统的挖掘效率和准确性。 本文为实现以人为中心的探索性数据挖掘及提高其实用性作了初步的探索工作。更多还原

【Abstract】 Today, the technology of data mining abstracting valuable information and running rule from a great deal data has its important economic value and academic significance, for whether business management or corporation decision-making. The technology of database has developed for 30 years, but there are many unsolved problems in the new field, so it is in the period that the created scholars put forward new concepts and the researchers resolve the problems. In current market, there are many kinds of all-purpose data mining systems applied commercial mode, but in fact almost nobody can operate these systems very well, only the people who are familiar with the technology of data mining can understand and use them, it is difficult to apply these technologies to solve their questions for common user.Because the traditional data-mining model lacks the means that express sufficiently the mining object and necessary interaction, and the standard of estimate is singleness, it cannot meet the need of the current application. Though the designed efficient arithmetic and optimized strategy can relieve the conflict in a way, it cannot really resolve the problem. The paper thinks about the user sufficiently and attempts to discuss the resolved ways from the aspect about the whole structure of data mining system, its stress is to bring forward the data mining system model based meta-knowledge and design flexible arithmetic.The paper introduces mainly three important properties of data mining system: Interaction, Control, and Usability. If the user do not know the field of data mining, the type of information that he will mine, and how to express the ideas in the system and the database that he will meet, the paper can help you greatlyunderstand the interrelated questions very well through metaknowledge. The paper proposes the content of meta-knowledge, organization and application model according with the standard of meta-data exchange, and designs a data mining system structure based on meta-knowledge which really confuses database and data mining. The meta-knowledge can provide some ways when the domain knowledge is organized, expressed and used, while monitor mining procession which can scale the amount of data and acquire enough bases, can diversify the source of data and obtain criterion and control.Association rule is an important content in data mining, efficient arithmetic is the most important part of data-mining 昺odel, especially when facing the complicated and various demands, it is necessary to propose characteristic and complementary arithmetic according to different factors. The paper analyses elaborately the idea of association rule mining arithmetic based on constraints, studies the classes of constraints in order to confirm its utility. The arithmetic aiming at united variable constraints improvement strategy and exchange of complication aggregation constraints has been designed, the practice has proved that it is effective. And then some concept such as support match mode, 0 support constraint and matching principle have been given, further more supporting the alterable support arithmetic based on enumerate mode tree that gives smallest support has been designed too. The meta-knowledge provides agility means for realizing the above arithmetic.At last, the paper discusses the function of user interface and gives the method of selecting constraints according to information plus, the operating flow of association rule mining system based on meta-knowledge, and the idea of inducting meta-knowledge in theevaluating phase in order to unit evaluating rule to whole procession, thus can do better in enhancing mining efficiency and veracity.The paper do some thing which does it best at searching after primarily how to realize the exploring data-mining system oriented user and enhance its utility.更多还原